Introduction

In this report, the biology of sars-cov-2 is discussed and the sars-cov-2 genome is analyzed. Machine learning will be used predict the translation transformations and deep learning will be used to predict the next nucleotide in the genome sequence.

SARS-CoV-2 Biology

Coronaviruses are a sphere-like 20-sided polygon, icosahedron, shape and belong to the Nidovirales order. SARS-CoV-2 travels through the respitory tract and attaches itself to alveoli, air sacs, in the lungs. Coronaviruses are named after the dyed red spike glycoproteins (S) that cover them and attach these protiens by being activated by the host cell's proteases, ACE2 and TMPRSS2 ennzymes, at a sequential proteolytic cleavage S1 and S2 site to fuse with a host cell's enzymes to enter, endocytosis, the cell's plasma membrane to the cytosol. S1 is the end binding region of the S protein and S2 is the beam of the S protien. The spike glcoproteins can also attach to a cell's endosomes being activated by the host cell's CTSL enzyme. Inside of the icosahedron positive-sensed lipid bilayer envelope is a helical capsid that encloses an RNA genome (gRNA), which encodes the viral genetic information -- non-segmented-single-stranded RNA -- that is used during the lytic cycle to generate RNA polymerase, protease, and other proteins. The envelope is removed allowing genomic RNA to enter the cytoplasm where some of it is translated into pp1a and pp1ab proteins. Translation is when codons, groups of three nucleotides, move along a ribosome through tRNA that translates the codons to anticodons that generate an amino acid polypeptide chain of proteins. Both proteins are then cleaved by protease making 16 nonstructural proteins, NSPs. Some of the NSPs form a replication and transcription complex (RTC) in which gRNA is replicated and RNA-dependent RNA polymerase (RdRp) uses gRNA as a template to transcribe (-) subgenomic RNA (sgRNA) and then transcribe it back to (+) sgRNA, which is used as the genome of the new virus. sgRNA is translated into strutural proteins, wich are spike protein (S), envelope protein (E), membrane protein (M), and nucleocapsid protein (N). Nucleocapsid proteins assign the viral replication-transcription complexes (RTCs), where gRNA and sgRNAs are synthesized. Spike, envelope, and membrane proteins enter the endoplasmic reticulum, and the nucleocapsid protein encloses the sgRNA becoming a nucleoprotein complex. The proteins and nucleoprotein complex combine into a virus endoplasmic reticulum-Golgi apparatus, and then are excreted into a vesicle. The dyed yellow envelope proteins (E) and dyed orange membrane glycoproteins (M) assist in budding -- using the host cell's plasma membrane for envelope formation -- and then exiting, exocytosis, the host cell through lysis. The following is a diagram of the SARS-CoV-2 lifecycle,

Sars-cov-2 has the largest RNA viral genome, which ranges from 26 to 32 kb. The genome of the virus starts with 5' carbon base, has an ORF1ab, S, E, M, N, and accesory region, and then end with 3' carbon base. ORF1ab are non-structural protiens. ORF1ab, open reading frames, comprises of 16 NSP regions. Nsp1 and nsp2 encode enzymes that bind to RNA and suppress gene expression. Nsp3 encodes enzymes that assist in the translation of viral mRNAs. Nsp4 encodes viroporin, which modify cellular membranes. Nsp5 encodes the organization of the viroplasm, where viral replication and assembly occur. Nsp6 generates autophagosomes, vesicles with cellular material that will be removed through autophagy. Nsp7 through 16 encode the generation of RNA synthesis and processing. Nsp7 through 11 encode the contribution to the nsp interactome, molecular interactions. Nsp12 encodes RNA polymerase. Nsp13 encodes helicase. Nsp14 through 16 encode mRNA capping, attaches the 5' cap to mRNA. The S region encodes the spike proteins, which bind to host cell receptors and fuse to the cell membrane. The M and E regions encode the membrane and envelope protiens, which assist in budding. The N region encodes the nucleocapsid protein, which assigns the viral replication-transcription complexes. The genome additionally contains an ORF coding for accessory proteins, which are not essential in virus replication but assist in pathogenesis, development of the virus.

ACE2 modulates the activity of the angiotensin II protein, which increases blood pressure and inflammation. The SARS-CoV-2 virus binding to ACE2 inhibits ACE2 and can cause an increase in cell inflammation resulting in the death of cells in the alveoli units with hypoxic fluid flooding, an increase in blood preasure damaging blood vessels causing microvascular thrombosis, and the tranduction of the kidney podocytes resulting in acute kidney injury. The following is a diagram of the enzymes that get inhibited and the parts of the body that can get damaged.

Testing for SARS-CoV-2

A patient can be tested for SARS-CoV-2 with the sequence-specific molecular nucleic acid assay, or the antigen-specific immunoassay. For a molecular-assay, a nasal or saliva sample is collected from upper respiratory fluid and then Real-time RT-PCR is conducted to quantify sequences within the RNA samples. Reverse transcriptase converts extracted RNA into cDNA that is used as a template for DNA polymerase to complete the strand of dsDNA and then RT-PCR amplifies the genetic regions and fluorescent probes bind to the regions for identification. For an immuno-assay, antibodies are put on a membrane and complexed with a potentialy virulent sample of which any antigen will be trapped that results in the membrane changing color. A SARS-CoV-2 recovered patient can be tested for whether the patient has developed anibodies against the virus by checking a blood sample for the antibodies. The following are diagrams of a molecular assay

and an immuno assay.

Vaccines for SARS-CoV-2

Vaccines have been developed to combat SARS-CoV-2. The Pfizer/BioNTech vaccine (BNT162b2) and The ModernaTX, Inc vaccine (mRNA-1273) are mRNA vaccines. mRNA vaccines deliver lab designed viral mRNA made by coding both the 5’-untranslated region (UTR) and the 3’-UTR into cells where it is translated into the encoded antigen to which an immune recognition by white blood cells, immunogenicity, results. The AstraZeneca in collaboration with the University of Oxford vaccine (AZD1222) is an adenovirus vaccine. Adenovirus vaccines are first gene sub-cloned into an intermediary vector, transfered to an adenovirus recombinant genome vector, transfected into packaging cells, amplified into a culture stock, and then titrated to determine the concentration of active adenoviruses in the stock. The adenoviruse is then delivered to cells where immunogenicity results. The following is a diagram of Virus vector DNA and mRNA vaccine immune response production. Vaccines have been developed to combat SARS-CoV-2. The Pfizer/BioNTech vaccine (BNT162b2) and The ModernaTX, Inc vaccine (mRNA-1273) are mRNA vaccines. mRNA vaccines deliver lab designed viral mRNA made by coding both the 5’-untranslated region (UTR) and the 3’-UTR into cells where it is translated into the encoded antigen to which an immune recognition by white blood cells, immunogenicity, results. The AstraZeneca in collaboration with the University of Oxford vaccine (AZD1222) is an adenovirus vaccine. Adenovirus vaccines are first gene sub-cloned into an intermediary vector, transfered to an adenovirus recombinant genome vector, transfected into packaging cells, amplified into a culture stock, and then titrated to determine the concentration of active adenoviruses in the stock. The adenoviruse is then delivered to cells where immunogenicity results. The following is a diagram of Virus vector DNA and mRNA vaccine immune response production.

Import data of sars-cov-2 genome.

Translation has three steps: initiation, which is when a ribosome, mRNA (sgRNA is a type of mRNA), and tRNA connect; elongation, which is when tRNA brings amino acids to the ribosome to be linked; and termination, which is when the polypeptide chain is complete. In this report, machine learning is used in two steps to complete the elongation process. First, the alogorithm will bind the codons to anticodons and then it will transfer the anticodons to amino acids.

Data Mining

The genome a '\n' character that is python for new line that will be removed.

Data Cleaning

Remove '\n' from genome and protein strings.

Makes a dataframe of sars-cov-2 mRNA nucleotides and mRNA antinucleotides.

Translate codons into corresponding anticodons.

Data Exploration

Sars-cov-2 genome is made up of apoximatly 30% adenine, 18% guanine, 20% cytosine and 32% uracil.

Percentage of uricil in genome data.

Percentage of guanine in genome data.

Percentage of adonine in genome data.

Percentage of cytosine in genome data.

There is a class imbalance that will be resolved by oversampling.

Feature Engineering

Represent nucleotides with numbers.

Oversampling selects data from minority class population to make up a larger share of the sample.

Reshape X and y so that they are 2d.

Resample independent and dependent variable.

Make a train test split of 80/20.

Machine Learning Modeling

The following models will be used.

Decision Trees split a node into two or more sub-nodes until node impurity is reduced and information gain is maximized.

Fit the model to the training data.

Show hyperparameters.

The algorithm uses the gini index to reduce impurity during desicion splits. c is the number of classes and p is the probability of the class.

$$ Gini=1-∑_{i}^{c}(p_{i})^{2}$$

Make predictions.

All predictions were classified correctly.

Seperate actual and predicted values into each class for metrics.

$$ precision = \frac{TP}{TP+FP} $$
$$ recall=\frac{TP}{TP+FN} $$
$$ f1=\frac{2*precision*recall}{precision+recall} $$
$$ accuracy=\frac{TP+TN}{TP+FN+TN+FP} $$
$$ specificity=\frac{TN}{TN+FP} $$
$$ True Positive Rate=\frac{TP}{TP+FN} $$
$$ False Positive Rate=\frac{FP}{FP+TN} $$
$$ Receiver Operating Characteristic=\frac{TPR}{FPR} $$
$$ Area Under Curve=∫_{a}^{b}TPR(FPR^{-1}(x))dx $$

Random forests are an ensemble of decision trees that output the average prediction of the trees.

Fit the model to the training data.

Show hyperparameters.

Make predictions.

All predictions were classified correctly.

Seperate predicted values into each class for metrics.

Gradient boosting is an ensemble of weak learning decision trees that iteratively combine into a single strong learner. Each tree is combined with the previous one in order to minimize the overall prediction error by optimizing a differentiable loss function. f0(x) is the initial tree, m is the iteration, ϕ𝑚(𝑥) is the current tree, and θ𝑚 is the weight for iteration m.

$$ f(x)=f_{0}(x)+∑_{m}^{M}θ_{m}ϕ_{m}(x) $$

Fit the model to the training data.

Show hyperparameters.

The direction of the steepest descent is given by the negative gradient, −𝑔(𝑥), of the loss function, L. At each iteration, a tree is fitted to predict the negative gradient for the next tree, ϕ𝑚. The negative gradient determines the direction of the step and line search, proposed by Friedman, determines the step length of ρ with an η learning rate.

$$ L_{MSE}= \frac{1}{N}∑_{i}^{N}(y_{i}-\hat{y}_{i})^{2} $$
$$ -g_{m}(x)=-∇\frac{∂L_{MSE}}{∂f(x)} $$
$$ ϕ_{𝑚}=argmin∑_{i}^{N}[(-g_{m}(x_{i}))-ϕ_{m-1}(x_{i})]^{2} $$
$$ ρ_{m}=argmin∑_{i}^{N}[((-g_{m}(x_{i}))+ϕ_{𝑚}(x_{i}))-y_{i}]^{2} $$
$$ f(x)=f_{0}(x)+∑_{m}^{M}ηρ_{m}ϕ_{m}(x)$$

Make predictions.

All predictions were classified correctly.

Seperate predicted values into each class for metrics.

All models performed with a cross validated roc auc score of 1. Will test on gradient boost model.

Fit the model to the test data.

Make predictions.

All predictions were classified correctly.

Seperate actual and predicted values into each class for metrics.

Next transfer anticodons to amino acids.

Data Cleaning

Take predictions from previous model and change them back to genome letters.

Break anticodons into threes.

Make a dataframe of anticodons and amino acids.

Remove extra nucleotide.

Anticodon conversion to amino acid chart.

Amino acids chart.

Match anticodons with amino acids.

Data Exploration

There is a class imbalance that will be resolved with oversampling.

Feature Engineering

Represent anticodons and amino acids with numbers.

Reshape X and y so that they are 2d.

Resample independent and dependent variable.

Make a train test split of 80/20.

Machine Learning Modeling

Fit the model to the training data.

Make predictions.

All predictions were classified correctly.

Seperate actual and predicted values into each class for metrics.

Fit the model to the training data.

Make predictions.

All predictions were classified correctly.

Seperate predicted values into each class for metrics.

Fit the model to the training data.

Make predictions.

All predictions were classified correctly.

Seperate predicted values into each class for metrics.

All models performed with a cross validated roc auc score of 1. Will test on gradient boost model.

Fit model to test data.

Make predictions.

All predictions were classified correctly.

Seperate actual and predicted values into each class for metrics.

The following is an rnn that is under development.

Data Cleaning

Adds a lagged RNA nucleotides column for nucleotide sequence prediction.

Data Exploration

Feature Engineering

XOR

Deep learning Modeling